slide2vec 1.2.3__tar.gz → 1.3.0__tar.gz

{slide2vec-1.2.3/slide2vec.egg-info → slide2vec-1.3.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: slide2vec
-Version: 1.2.3
+Version: 1.3.0
 Summary: Embedding of whole slide images with Foundation Models
 Home-page: https://github.com/clemsgrs/slide2vec
 Author: Clément Grisi

{slide2vec-1.2.3 → slide2vec-1.3.0}/pyproject.toml

@@ -23,7 +23,7 @@ warn_unused_configs = true
 no_implicit_reexport = true
 
 [tool.bumpver]
-current_version = "1.2.3"
+current_version = "1.3.0"
 version_pattern = "MAJOR.MINOR.PATCH"
 commit = false       # We do version bumping in CI, not as a commit
 tag = false          # Git tag already exists — we don't auto-tag

{slide2vec-1.2.3 → slide2vec-1.3.0}/setup.cfg

@@ -1,6 +1,6 @@
 [metadata]
 name = slide2vec
-version = 1.2.3
+version = 1.3.0
 description = Embedding of whole slide images with Foundation Models
 author = Clément Grisi
 platforms = unix, linux, osx, cygwin, win32

slide2vec-1.3.0/slide2vec/__init__.py

@@ -0,0 +1 @@
+__version__ = "1.3.0"

{slide2vec-1.2.3 → slide2vec-1.3.0}/slide2vec/aggregate.py

@@ -33,6 +33,9 @@ def get_args_parser(add_help: bool = True):
         default="",
         help="Name of output subdirectory",
     )
+    parser.add_argument(
+        "--run-on-cpu", action="store_true", help="run inference on cpu"
+    )
     return parser
 
 
@@ -49,6 +52,7 @@ def scale_coordinates(wsi_fp, coordinates, spacing, backend):
 
 def main(args):
     # setup configuration
+    run_on_cpu = args.run_on_cpu
     cfg = get_cfg_from_file(args.config_file)
     output_dir = Path(cfg.output_dir, args.run_id)
     cfg.output_dir = str(output_dir)
@@ -87,7 +91,7 @@ def main(args):
 
     autocast_context = (
         torch.autocast(device_type="cuda", dtype=torch.float16)
-        if cfg.speed.fp16
+        if (cfg.speed.fp16 and not run_on_cpu)
         else nullcontext()
     )
     feature_aggregation_updates = {}
@@ -136,7 +140,8 @@ def main(args):
 
             torch.save(wsi_feature, feature_path)
             del wsi_feature
-            torch.cuda.empty_cache()
+            if not run_on_cpu:
+                torch.cuda.empty_cache()
             gc.collect()
 
             feature_aggregation_updates[str(wsi_fp)] = {"status": "success"}

{slide2vec-1.2.3 → slide2vec-1.3.0}/slide2vec/embed.py

@@ -33,6 +33,9 @@ def get_args_parser(add_help: bool = True):
         default="",
         help="Name of output subdirectory",
     )
+    parser.add_argument(
+        "--run-on-cpu", action="store_true", help="run inference on cpu"
+    )
     return parser
 
 
@@ -61,14 +64,15 @@ def create_dataset(wsi_fp, coordinates_dir, spacing, backend, transforms):
     )
 
 
-def run_inference(dataloader, model, device, autocast_context, unit, batch_size, feature_path, feature_dim, dtype):
+def run_inference(dataloader, model, device, autocast_context, unit, batch_size, feature_path, feature_dim, dtype, run_on_cpu: False):
+    device_name = f"GPU {distributed.get_global_rank()}" if not run_on_cpu else "CPU"
     with h5py.File(feature_path, "w") as f:
         features = f.create_dataset("features", shape=(0, *feature_dim), maxshape=(None, *feature_dim), dtype=dtype, chunks=(batch_size, *feature_dim))
         indices = f.create_dataset("indices", shape=(0,), maxshape=(None,), dtype='int64', chunks=(batch_size,))
         with torch.inference_mode(), autocast_context:
             for batch in tqdm.tqdm(
                 dataloader,
-                desc=f"Inference on GPU {distributed.get_global_rank()}",
+                desc=f"Inference on {device_name}",
                 unit=unit,
                 unit_scale=batch_size,
                 leave=False,
@@ -86,7 +90,8 @@ def run_inference(dataloader, model, device, autocast_context, unit, batch_size,
                 del image, feature
 
     # cleanup
-    torch.cuda.empty_cache()
+    if not run_on_cpu:
+        torch.cuda.empty_cache()
     gc.collect()
 
 
@@ -116,11 +121,13 @@ def load_sort_and_deduplicate_features(tmp_dir, name, expected_len=None):
 
 def main(args):
     # setup configuration
+    run_on_cpu = args.run_on_cpu
     cfg = get_cfg_from_file(args.config_file)
     output_dir = Path(cfg.output_dir, args.run_id)
     cfg.output_dir = str(output_dir)
 
-    setup_distributed()
+    if not run_on_cpu:
+        setup_distributed()
 
     if cfg.tiling.read_coordinates_from:
         coordinates_dir = Path(cfg.tiling.read_coordinates_from)
@@ -155,7 +162,8 @@ def main(args):
         model = ModelFactory(cfg.model).get_model()
         if distributed.is_main_process():
             print(f"Starting {unit}-level feature extraction...")
-        torch.distributed.barrier()
+        if not run_on_cpu:
+            torch.distributed.barrier()
 
         # select slides that were successfully tiled but not yet processed for feature extraction
         tiled_df = process_df[process_df.tiling_status == "success"]
@@ -174,7 +182,7 @@ def main(args):
 
         autocast_context = (
             torch.autocast(device_type="cuda", dtype=torch.float16)
-            if cfg.speed.fp16
+            if (cfg.speed.fp16 and not run_on_cpu)
             else nullcontext()
         )
         feature_extraction_updates = {}
@@ -231,9 +239,11 @@ def main(args):
                     tmp_feature_path,
                     feature_dim,
                     dtype,
+                    run_on_cpu,
                 )
 
-                torch.distributed.barrier()
+                if not run_on_cpu:
+                    torch.distributed.barrier()
 
                 if distributed.is_main_process():
                     wsi_feature = load_sort_and_deduplicate_features(tmp_dir, name, expected_len=len(dataset))
@@ -241,10 +251,12 @@ def main(args):
 
                     # cleanup
                     del wsi_feature
-                    torch.cuda.empty_cache()
+                    if not run_on_cpu:
+                        torch.cuda.empty_cache()
                     gc.collect()
 
-                torch.distributed.barrier()
+                if not run_on_cpu:
+                    torch.distributed.barrier()
 
                 feature_extraction_updates[str(wsi_fp)] = {"status": "success"}
 

{slide2vec-1.2.3 → slide2vec-1.3.0}/slide2vec/main.py

@@ -18,6 +18,12 @@ def get_args_parser(add_help: bool = True):
     parser.add_argument(
         "--config-file", default="", metavar="FILE", help="path to config file"
     )
+    parser.add_argument(
+        "--skip-datetime", action="store_true", help="skip run id datetime prefix"
+    )
+    parser.add_argument(
+        "--run-on-cpu", action="store_true", help="run inference on cpu"
+    )
     return parser
 
 
@@ -59,7 +65,7 @@ def run_tiling(config_file, run_id):
         sys.exit(proc.returncode)
 
 
-def run_feature_extraction(config_file, run_id):
+def run_feature_extraction(config_file, run_id, run_on_cpu: False):
     print("Running embed.py...")
     # find a free port
     with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
@@ -77,6 +83,16 @@ def run_feature_extraction(config_file, run_id):
         "--config-file",
         config_file,
     ]
+    if run_on_cpu:
+        cmd = [
+            sys.executable,
+            "slide2vec/embed.py",
+            "--run-id",
+            run_id,
+            "--config-file",
+            config_file,
+            "--run-on-cpu",
+        ]
     # launch in its own process group.
     proc = subprocess.Popen(
         cmd,
@@ -103,7 +119,7 @@ def run_feature_extraction(config_file, run_id):
         sys.exit(proc.returncode)
 
 
-def run_feature_aggregation(config_file, run_id):
+def run_feature_aggregation(config_file, run_id, run_on_cpu: False):
     print("Running aggregate.py...")
     # find a free port
     cmd = [
@@ -114,6 +130,8 @@ def run_feature_aggregation(config_file, run_id):
         "--config-file",
         config_file,
     ]
+    if run_on_cpu:
+        cmd.append("--run-on-cpu")
     # launch in its own process group.
     proc = subprocess.Popen(
         cmd,
@@ -131,7 +149,7 @@ def run_feature_aggregation(config_file, run_id):
             sys.stdout.flush()
         proc.wait()
     except KeyboardInterrupt:
-        print("Received CTRL+C, terminating embed.py process group...")
+        print("Received CTRL+C, terminating aggregate.py process group...")
         os.killpg(os.getpgid(proc.pid), signal.SIGTERM)
         proc.wait()
         sys.exit(1)
@@ -142,18 +160,18 @@ def run_feature_aggregation(config_file, run_id):
 
 def main(args):
     config_file = args.config_file
+    skip_datetime = args.skip_datetime
+    run_on_cpu = args.run_on_cpu
 
-    cfg = setup(config_file)
+    cfg, run_id = setup(config_file, skip_datetime=skip_datetime)
     hf_login()
 
-    output_dir = Path(cfg.output_dir)
-    run_id = output_dir.stem
-
     run_tiling(config_file, run_id)
 
     print("Tiling completed.")
     print("=+=" * 10)
 
+    output_dir = Path(cfg.output_dir)
     features_dir = output_dir / "features"
     if cfg.wandb.enable:
         stop_event = threading.Event()
@@ -162,15 +180,17 @@ def main(args):
         )
         log_thread.start()
 
-    run_feature_extraction(config_file, run_id)
-    print("Feature extraction completed.")
-    print("=+=" * 10)
+    run_feature_extraction(config_file, run_id, run_on_cpu)
 
     if cfg.model.level == "slide":
-        run_feature_aggregation(config_file, run_id)
-        print("Feature aggregation completed.")
+        run_feature_aggregation(config_file, run_id, run_on_cpu)
+        print("Feature extraction completed.")
+        print("=+=" * 10)
+    else:
+        print("Feature extraction completed.")
         print("=+=" * 10)
 
+
     if cfg.wandb.enable:
         stop_event.set()
         log_thread.join()
@@ -180,16 +200,21 @@ def main(args):
 
 
 if __name__ == "__main__":
-    import warnings
 
+    import warnings
     import torchvision
+    
     torchvision.disable_beta_transforms_warning()
-
+    
     warnings.filterwarnings("ignore", message=".*Could not set the permissions.*")
     warnings.filterwarnings("ignore", message=".*antialias.*", category=UserWarning)
     warnings.filterwarnings("ignore", message=".*TypedStorage.*", category=UserWarning)
     warnings.filterwarnings("ignore", category=FutureWarning)
     warnings.filterwarnings("ignore", message="The given NumPy array is not writable")
+    warnings.filterwarnings(
+        "ignore",
+        message=".*'frozen' attribute with value True was provided to the `Field`.*"
+    )
 
     args = get_args_parser(add_help=True).parse_args()
     main(args)

{slide2vec-1.2.3 → slide2vec-1.3.0}/slide2vec/models/models.py

@@ -578,5 +578,5 @@ class PRISM(SlideFeatureExtractor):
     def forward_slide(self, tile_features, **kwargs):
         tile_features = tile_features.unsqueeze(0)
         reprs = self.slide_encoder.slide_representations(tile_features)
-        output = reprs["image_embedding"]  # [1, 1280]
+        output = reprs["image_embedding"].squeeze(0)  # [1280]
         return output

{slide2vec-1.2.3 → slide2vec-1.3.0}/slide2vec/utils/config.py

@@ -32,7 +32,7 @@ def get_cfg_from_file(config_file):
     return cfg
 
 
-def setup(config_file):
+def setup(config_file, skip_datetime: bool = False):
     """
     Basic configuration setup without any distributed or GPU-specific initialization.
     This function:
@@ -45,8 +45,10 @@ def setup(config_file):
 
     if cfg.resume:
         run_id = cfg.resume_dirname
-    else:
+    elif not skip_datetime:
         run_id = datetime.datetime.now().strftime("%Y-%m-%d_%H_%M")
+    else:
+        run_id = ""
 
     if cfg.wandb.enable:
         key = os.environ.get("WANDB_API_KEY")
@@ -56,7 +58,7 @@ def setup(config_file):
 
     output_dir = Path(cfg.output_dir, run_id)
     if distributed.is_main_process():
-        output_dir.mkdir(exist_ok=cfg.resume, parents=True)
+        output_dir.mkdir(exist_ok=cfg.resume or skip_datetime, parents=True)
     cfg.output_dir = str(output_dir)
 
     fix_random_seeds(0)
@@ -65,7 +67,7 @@ def setup(config_file):
     cfg_path = write_config(cfg, cfg.output_dir)
     if cfg.wandb.enable:
         wandb_run.save(cfg_path)
-    return cfg
+    return cfg, run_id
 
 
 def setup_distributed():

{slide2vec-1.2.3 → slide2vec-1.3.0/slide2vec.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: slide2vec
-Version: 1.2.3
+Version: 1.3.0
 Summary: Embedding of whole slide images with Foundation Models
 Home-page: https://github.com/clemsgrs/slide2vec
 Author: Clément Grisi

slide2vec-1.2.3/slide2vec/__init__.py

@@ -1 +0,0 @@
-__version__ = "1.2.3"